---
kind: concept
slug: location-encoder
title: Location encoder
summary: >-
  Maps latitude and longitude to a fixed-length embedding for downstream models instead of raw coordinates. Variants: raw coordinates, or a learned encoder (spherical harmonics plus a SIREN MLP) pretrained by contrastive matching to satellite tiles, where the degree parameter L trades held-out-continent generalisation against regional accuracy.
tags: [geospatial, location-encoding, pretraining]
papers: ["2311.17179"]
related: [contrastive-learning, implicit-neural-representation, positional-encoding, fourier-features, multimodal-fusion]
status: draft
updated_at: 2026-11-09
---

## Overview

A location encoder turns a longitude and latitude pair into a vector that a
downstream model can use in place of the raw numbers. Raw coordinates
generalise poorly: a model built on them tends to work only where the
evaluation area overlaps the training area [arxiv:2311.17179 §1]. Image-based
alternatives avoid this but need a satellite image and a vision model for every
query point [arxiv:2311.17179 §2].

The idea shared by the learned variants is to pretrain the encoder so that
places with similar ground appearance get similar embeddings, then freeze it
and use its output as features. The invariant that holds across the variants is
that prediction needs only the coordinates, not imagery
[arxiv:2311.17179 §2].

Only one paper with a card covers this concept so far, so the variants below
all come from that paper. Other encoders it compares against are named only.

## Variants

### Raw coordinates

Feed longitude and latitude straight into the downstream model. The paper uses
this as its "Identity" baseline [arxiv:2311.17179 tab. 2]. It needs no
pretraining and no imagery.

### Spherical harmonics with a SIREN MLP, pretrained against satellite tiles

The coordinates are expanded into real spherical-harmonic basis functions up to
a degree limit L. A two-hidden-layer SIREN network with 512-unit hidden layers
maps this to a 256-dimensional embedding [arxiv:2311.17179 §2.2]
[arxiv:2311.17179 app. C.1]. Training matches each coordinate with the
satellite tile at that place, using a CLIP-style contrastive loss against a
frozen MoCo image encoder whose last linear projection is trained
[arxiv:2311.17179 §2.1] [arxiv:2311.17179 §2.2]. Card:
[SatCLIP](/papers/2311.17179).

- **L = 10.** The authors report this setting as the better generaliser to a
  continent held out from training [arxiv:2311.17179 §4.2].
- **L = 40.** A finer basis. The authors report it as stronger on
  in-distribution regional tasks and on interpolation [arxiv:2311.17179 §4.1]
  [arxiv:2311.17179 §4.4].

The paper's released checkpoints are the L = 10 and L = 40 models
[arxiv:2311.17179 §2.2].

### Other pretrained location encoders

CSP, GPS2Vec, GeoCLIP and MOSAIKS are the comparison methods in the paper
[arxiv:2311.17179 §3.3] [arxiv:2311.17179 tab. 1]. They have no cards here, so
this page does not describe how they work.

## When to use which

The rows follow the constraints. The results are the paper's own, on its own
tasks, and the authors report them.

| Situation | Use | Why |
|---|---|---|
| Test points are spread like the pretraining data, e.g. random splits over the globe | SatCLIP, L = 40 | Best on seven of nine in-distribution tasks, with L = 40 scoring higher than L = 10 on the regional tasks [arxiv:2311.17179 tab. 2] [arxiv:2311.17179 §4.1]. GeoCLIP is equal or better on two of them. |
| Downstream data lies on a continent left out of pretraining, zero-shot | SatCLIP, L = 10 | L = 10 has more winning rows than L = 40 on the held-out tests [arxiv:2311.17179 tab. 3]. It is not best everywhere: MOSAIKS and GeoCLIP win several rows [arxiv:2311.17179 tab. 3]. |
| Held-out continent, with 1% of its points available as labels | SatCLIP, L = 10 | On the four few-shot rows for Asia and Africa (Countries and Ecoregions), L = 10 scores higher than L = 40 [arxiv:2311.17179 tab. 3]. |
| Only coordinates are available and no pretrained encoder can be used | Raw coordinates | Cheapest option, but the Identity baseline is below both SatCLIP settings on all nine in-distribution rows [arxiv:2311.17179 tab. 2]. |
| Must train your own encoder | Expect a large job | The paper uses about 100,000 paired tiles, a batch of 8k, 500 epochs and one A100 for about two days [arxiv:2311.17179 §3.1] [arxiv:2311.17179 §2.3] [arxiv:2311.17179 app. C.1]. Reuse a released checkpoint if its terms allow. |

Two points limit these choices. Table 2 and Table 3 use different image
encoders (ResNet50 and ViT16), so compare numbers only within a table
[arxiv:2311.17179 tab. 2] [arxiv:2311.17179 tab. 3]. The paper also reports
that the vision encoder changes results by under 1%, while L changes them more
[arxiv:2311.17179 §4.4].

## Implementation sketch

Language-agnostic outline of the learned encoder's forward pass, following the
paper's description. Values the paper leaves open are marked.

```pseudocode
# Learned location encoder  [arxiv:2311.17179 §2.2]
# Input:  coords (N, 2) as (longitude, latitude) in degrees
# Output: embedding (N, 256)

def location_encoder(coords, L):
    # Real spherical-harmonic basis, degrees below L  [arxiv:2311.17179 §2.2]
    basis = spherical_harmonic_features(coords, L)     # (N, basis size)

    # First SIREN layer: linear, then sine  [arxiv:2311.17179 app. C.1]
    # w0 is the sine frequency; the paper does not state it, see Contradictions
    h = sin(w0 * (basis @ W1 + b1))

    # Second hidden layer, 512 units in the paper  [arxiv:2311.17179 app. C.1]
    h = sin(h @ W2 + b2)

    # Linear map to the 256-dimensional output  [arxiv:2311.17179 §2.2]
    return h @ W3 + b3
```

Pretraining objective, one batch of N coordinate and tile pairs
[arxiv:2311.17179 §2.1]:

```pseudocode
# Symmetric contrastive loss over one batch  [arxiv:2311.17179 §2.1]
# The true pair for row i is column i

def contrastive_loss(tile_emb, loc_emb, tau):
    # Unit-normalise both sides, so dot products are cosines
    tile_emb = tile_emb / norm(tile_emb)
    loc_emb = loc_emb / norm(loc_emb)
    # N x N similarity matrix, scaled by the temperature tau
    logits = (tile_emb @ loc_emb.T) / tau
    # Each tile against all coordinates, and each coordinate against all tiles
    # tau is a hyperparameter; the paper gives no number  [arxiv:2311.17179 §2.1]
    return (cross_entropy(logits, arange(N))
            + cross_entropy(logits.T, arange(N))) / 2
```

Notes:

- Input order matters. The encoder expects longitude first and latitude second
  in degrees. Swapping the columns gives a valid but wrong embedding.
- The batch size matters for the contrastive loss. The paper uses 8k, so each
  coordinate is compared against 8k tiles per step [arxiv:2311.17179 §2.3].
- The released code differs from the paper in several settings, listed under
  Contradictions.

## Contradictions

Each side is cited. None of these are resolved here.

- **Batch size.** The paper trains with 8k [arxiv:2311.17179 §2.3]. The
  repository's default config uses 512 with 16 accumulation steps, which gives
  512 negatives per loss computation. See the
  [SatCLIP card, Paper vs code](/papers/2311.17179).
- **SIREN width.** The paper uses 512 hidden units [arxiv:2311.17179 app. C.1].
  The repository's default is 256 (same card).
- **Degree limit.** The paper releases L = 10 and L = 40 [arxiv:2311.17179 §2.2].
  The repository's default is 32 (same card).
- **Image encoder.** The paper uses a frozen MoCo encoder with only the last
  projection trained [arxiv:2311.17179 §2.2]. The repository's default config
  builds a smaller ViT from scratch with no freezing (same card).
- **ViT16 backbone.** The paper gives about 22M parameters for ViT16
  [arxiv:2311.17179 §2.2]. The repository's README names ViT-B/16 and its code
  uses a ViT-Small (same card).
- **Sine frequency and temperature.** The paper gives no number for the sine
  frequency or the temperature τ [arxiv:2311.17179 §2.1]. The repository's code
  sets a specific first-layer frequency and learns τ (same card).
- **Dataset licence.** The paper does not state a licence for its satellite
  tile dataset [arxiv:2311.17179 app. A]. Check it before reusing the tiles
  commercially. The paper's own CC BY 4.0 licence covers only the paper.
- **Author claims.** The authors report the encoder best on seven of nine
  in-distribution tasks [arxiv:2311.17179 §4.1]. Table 2 has GeoCLIP equal or
  ahead on the two income and housing tasks [arxiv:2311.17179 tab. 2]. All
  results are author-reported, and the released checkpoints have not been
  checked against the paper here (see the card).

No contradictions with other cards are known, since this is the only card so far.

## Papers

- [SatCLIP: Global, General-Purpose Location Embeddings with Satellite Imagery](/papers/2311.17179): contrastive pretraining of a spherical-harmonic and SIREN location encoder against Sentinel-2 tiles, with L = 10 and L = 40 models.